Reconstruct what happened before committing to an explanation.

Gap Last is a way to understand something you observed, were told, or otherwise came to be aware of. Something arrived. You did not get to rerun it. A story is already forming.

The formal method is constraint-first reconstruction. You can say "run this through Gap Last" without reading a paper. A working draft of that paper is here if you want the longer argument.

You cannot answer how until you have a solid what. How is a mechanism. What is the object. A how aimed at a thin what is a story with better vocabulary.

Pursue the how at your folly if you fail to frame the correct what.

A parent, a patient, a founder, and an agent all face the same situation: something arrived, you did not get to rerun it, and a story is already forming.

Gap Last got its name from a week in late August 2026, watching the understanding of a flood off Langtang Lirung move from ice to rock. The Gap Last method is what happened there, written down so it can continue to happen on purpose. You do not need that week to begin. A labeled ordinary incident is on the home page.

The name is Gap Last, not Last Gap. Naming a leftover question is not the same as knowing you have the last one. A later trace can move the question. Treating the gap you named as closed, when you do not yet know enough to say so, is the same thin move the method exists to stop.

It exists because a cause that is wrong, or too thin, aims the next warning system, the next treatment, the next fix at the wrong object. The people in the path of the next similar event deserve an accurate account. The events of that week on Langtang Lirung sparked the method. A debris flood had buried homes downstream, and the first account, a glacier collapse, was already being defended and planned against while it was still only what the cloudy images could show. The next warning, the next wall, the next evacuation plan would have been built for ice. Thursday's images moved the failure to rock. A cause latched onto too early doesn't just miss. It spends the preparation on the wrong object, and the people below stay exposed.

If the first question is "who caused it?", reconstruct first. The prohibition is the product. The method does not promise that a causal account will always be available.

Chesterton's Fence is older, and related. It says find out why the fence stands before you clear it away. A fence is a cause somebody paid for, and tearing it down unread throws away what they paid. Gap Last stands on the other side of that fence: name the gap before you put one up, so the next person who finds your fence can read what it closed. Both refuse the same move, a cause written down ahead of the question.

If you can run the experiment, run it. If you can prove the theorem, prove it. If the bottleneck is sampling, do the statistics. Those are later steps, not the starting posture.

The order, at three lengths, is on the Gap Last method. The four-line exercise on the home page needs no install. To put the skill in an agent, see Install in your agent.

  • Method: the Gap Last method (formal name: constraint-first reconstruction)
  • Draft paper: constraint-first reconstruction
  • Tools: skill, CLI, reconstruction file
  • Primary start: skill ZIP or GitHub
  • npm gaplast: name hold as of 10 September 2026, not the implementation

Built by Catalyst Forge LLC. MIT.